Production process parameter dynamic adjustment method and device based on AI visual inspection

Abstract

The invention discloses a production process parameter dynamic adjustment method and device based on AI visual inspection. A server side method comprises the steps of obtaining a production video stream collected by a 5G explosion-proof camera pre-deployed in a production environment; inputting the production video stream into a pre-trained AI visual inspection model, and outputting a plurality of appearance feature parameters which correspond to the production video stream and are used for reflecting the appearance of the chemical product; according to the multiple appearance characteristic parameters, production process parameters which do not meet preset product production conditions in production equipment of the production environment are analyzed, and the parameters are obtained based on traceability of a preset process knowledge guiding graph; the preset process knowledge guide graph is used for recording a correlation relationship between the appearance process mark and a production process parameter set of production equipment in the production environment; and adjusting the parameter values of the production process parameters which do not meet the preset product production conditions. Therefore, by adopting the embodiment of the invention, the high requirement on quality control can be met, and meanwhile, the product error probability is reduced.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method and device for dynamically adjusting production process parameters based on AI visual inspection. Background Art

[0002] In the production of chemical products, especially in the fine chemical sector, product appearance characteristics (such as color, shape, and surface texture) are important indicators of quality and performance. In actual production, these appearance characteristics require real-time monitoring on the production line to enable timely adjustment of production process parameters to ensure consistent and stable product quality.

[0003] In the relevant technology, appearance inspection in chemical product production primarily relies on traditional testing equipment. Traditional testing equipment can typically only detect simple physical parameters of the product. Its accuracy for in-depth appearance features such as morphology and texture is low, making it difficult to meet the high demands of quality control. Furthermore, traditional testing equipment cannot adjust production equipment parameters in real time based on test results, increasing the probability of product errors. Summary of the Invention

[0004] The embodiments of this application provide a method and apparatus for dynamically adjusting production process parameters based on AI visual inspection. To provide a basic understanding of some aspects of the disclosed embodiments, a brief summary is provided below. This summary is not intended to be a comprehensive review, identify key or important elements, or delineate the scope of protection for these embodiments. Its sole purpose is to present some concepts in a simplified form, serving as a prelude to the detailed description that follows.

[0005] In a first aspect, an embodiment of the present application provides a method for dynamically adjusting production process parameters based on AI visual inspection, which is applied to a server. The method includes: The production video stream collected by the 5G explosion-proof camera pre-deployed in the production environment is obtained according to the preset cycle. The production video stream is used to represent the internal information of the reactor, material feeding information, and the appearance information of the chemical product; Input the production video stream into the pre-trained AI visual inspection model, and output multiple appearance feature parameters corresponding to the production video stream to reflect the appearance of chemical products; Analyze the production process parameters of the production equipment in the production environment that do not meet the preset product production conditions based on multiple appearance feature parameters; the production process parameters that do not meet the preset product production conditions are traced back based on the preset process knowledge-oriented graph, which is used to record the correlation between the appearance process marks and the production process parameter sets of the production equipment in the production environment; Adjust the parameter values ​​of production process parameters that do not meet the preset product production conditions according to the preset step size.

[0006] In a second aspect, an embodiment of the present application provides a device for dynamically adjusting production process parameters based on AI visual inspection, the device comprising: The production video stream acquisition module is used to obtain the production video stream collected by 5G explosion-proof cameras pre-deployed in the production environment within a preset time period. The production video stream is used to represent the internal information of the reactor, material feeding information, and the appearance information of the chemical product; The appearance feature parameter output module is used to input the production video stream into the pre-trained AI visual inspection model and output multiple appearance feature parameters corresponding to the production video stream that reflect the appearance of the chemical product; A production process parameter analysis module is used to analyze, based on multiple appearance feature parameters, production process parameters in production equipment in the production environment that do not meet the preset product production conditions. The production process parameters that do not meet the preset product production conditions are traced back based on a preset process knowledge-oriented graph, which is used to record the correlation between the appearance process mark and the set of production process parameters of the production equipment in the production environment; The parameter value adjustment module is used to adjust the parameter values ​​of production process parameters that do not meet the preset product production conditions according to the preset step size.

[0007] The technical solutions provided by the embodiments of the present application may have the following beneficial effects: In an embodiment of the present application, on the one hand, the pre-trained AI visual inspection model can extract multiple appearance feature parameters, including physical appearance parameters (such as color, shape) and non-physical appearance parameters (such as texture, gloss) from the production video stream collected by the 5G explosion-proof camera. These parameters can comprehensively reflect the appearance information of chemical products and provide more accurate and detailed data support for quality control. By analyzing these appearance feature parameters, any slight changes in the appearance of the product can be discovered in a timely manner, which can meet the high requirements for quality control. On the other hand, by obtaining multiple appearance feature parameters reflecting the appearance of chemical products in real time, these parameters are associated with the set of production process parameters recorded in the preset process knowledge-oriented graph to trace the non-compliant production process parameters. The system can automatically adjust the values ​​of these parameters to adapt to the current production conditions. The dynamic online adjustment mechanism can promptly correct deviations in the production process, reduce product quality problems caused by improper parameter settings, and thus significantly reduce the probability of product errors.

[0008] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0010] Figure 1 This is a schematic diagram of a method flow for dynamically adjusting production process parameters based on AI visual inspection provided by an embodiment of the present application; Figure 2 This is a schematic diagram of an image frame collected in an actual scene by a 5G explosion-proof camera provided in an embodiment of the present application; Figure 3 This is a schematic diagram of an application scenario provided by an embodiment of the present application; Figure 4 This is a model architecture diagram of a pre-trained AI visual detection model provided in an embodiment of the present application; Figure 5A This is a parameter representation of a physical appearance parameter provided in an embodiment of the present application; Figure 5B This is a parameter representation of a non-physical appearance parameter provided in an embodiment of the present application; Figure 6 This is a knowledge-oriented graph of a root node and multiple child nodes provided in an embodiment of the present application; Figure 7 This is a schematic diagram of part of the content of a preset process knowledge-oriented diagram provided in an embodiment of the present application; Figure 8 1 is a flow chart of a model training method for an action parameter analysis model provided in an embodiment of the present application; Figure 9 This is a structural diagram of a device for dynamically adjusting production process parameters based on AI visual inspection provided in an embodiment of the present application; Figure 10 This is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0011] The following description and the drawings sufficiently illustrate specific embodiments of the application to enable those skilled in the art to practice them.

[0012] It should be clear that the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0013] When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. Instead, they are merely examples of devices and methods consistent with certain aspects of the present application, as detailed in the appended claims.

[0014] In the description of this application, it should be understood that the terms "first", "second", etc. are used for descriptive purposes only and should not be understood as indicating or implying relative importance. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances. In addition, in the description of this application, unless otherwise specified, "multiple" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the previous and subsequent associated objects are in an "or" relationship.

[0015] At present, appearance inspection in chemical product production mainly relies on traditional inspection equipment.

[0016] The inventors realized that traditional testing equipment can only measure simple physical parameters of a product. Its accuracy for in-depth appearance features like shape and texture is low, making it difficult to meet the high demands of quality control. Furthermore, traditional testing equipment cannot adjust production equipment parameters in real time based on test results, increasing the probability of product errors.

[0017] To address the aforementioned issues, the present application provides a method and apparatus for dynamically adjusting production process parameters based on AI visual inspection, addressing the aforementioned related technical issues. In an embodiment of the present application, a pre-trained AI visual inspection model can extract multiple appearance feature parameters, including physical appearance parameters (such as color and shape) and non-physical appearance parameters (such as texture and gloss), from production video streams captured by 5G explosion-proof cameras. These parameters can comprehensively reflect the appearance of chemical products, providing more accurate and detailed data support for quality control. Analysis of these appearance feature parameters can promptly detect any minor changes in product appearance, meeting the high standards of quality control. Furthermore, by acquiring multiple appearance feature parameters reflecting the appearance of chemical products in real time and associating them with a set of production process parameters recorded in a pre-set process knowledge-based graph, the system can automatically adjust the values ​​of these parameters to adapt to current production conditions. This dynamic online adjustment mechanism can promptly correct deviations in the production process, reduce product quality issues caused by improper parameter settings, and significantly reduce the probability of product errors. This is explained in detail below using an exemplary embodiment.

[0018] The following will be combined with the Figure 1 -Attached Figure 8This article details the method for dynamically adjusting production process parameters based on AI visual inspection, as provided in an embodiment of this application. This method can be implemented using a computer program and run on a device for dynamically adjusting production process parameters based on AI visual inspection, which is based on a von Neumann architecture. This computer program can be integrated into an application or run as a standalone tool application.

[0019] See Figure 1 , provides a flow chart of a method for dynamically adjusting production process parameters based on AI visual inspection, which is applied to the server. Figure 1 As shown, the method of the embodiment of the present application includes the following steps: S101: Obtain, according to a preset cycle, a production video stream collected by a 5G explosion-proof camera pre-deployed in the production environment. The production video stream is used to represent the internal information of the reactor, material feeding information, and the appearance information of the chemical product; Among them, the preset cycle is a predetermined time interval for regularly performing dynamic adjustments to production process parameters. In this application, it specifically refers to the time interval for regularly obtaining video streams from the camera. The production video stream is video data captured and continuously transmitted from the camera in real time, which is used to monitor and analyze various information in the production process. The reactor is a commonly used container in chemical production for chemical reactions. Material feeding information is information such as the type, quantity and time of raw materials put into the reactor during the production process. The appearance information of chemical products refers to the physical appearance characteristics of chemical products, such as color, shape, etc. The 5G explosion-proof camera is an image acquisition device that can obtain clear video streams in harsh production environments. Faced with complex environments such as high temperature and high pressure in the reactor, there are certain difficulties in online real-time quality visualization detection and analysis of products. By deploying a 5G explosion-proof camera in the reactor, the camera has customized cooling backflush, circulating cooling water, support for variable focus mode, explosion-proof and other functions. The image frames collected by the camera in the actual production environment are, for example, Figure 2 shown.

[0020] In some embodiments of the present application, 5G explosion-proof cameras are deployed in key locations within the production environment, such as near reactors and at material feed ports. The cameras automatically activate and begin capturing production video streams. The server obtains the production video streams captured by the pre-deployed 5G explosion-proof cameras at preset intervals. This video stream is used to represent the internal information of the reactor, material feed information, and the appearance of the chemical product.

[0021] For example Figure 3 As shown in the figure, the 5G explosion-proof camera can collect production video streams that can represent the internal information of the reactor, material feeding information, and the appearance information of chemical products according to the preset cycle, and send the collected production video streams to the server through the 5G network.

[0022] S102: Input the production video stream into a pre-trained AI visual inspection model, and output multiple appearance feature parameters corresponding to the production video stream that reflect the appearance of the chemical product; Pre-trained AI visual inspection models are mathematical models that are trained using large amounts of labeled data, enabling them to recognize and understand features in videos. Multiple appearance feature parameters include physical and non-physical appearance parameters. Physical appearance parameters, such as color and shape, can be directly observed and measured. Non-physical appearance parameters, such as a product's complex texture, gloss, and transparency, cannot be directly observed and measured.

[0023] For example Figure 4 As shown, the pre-trained AI visual inspection model includes a basic appearance feature extraction module and a deep appearance feature extraction module; the basic appearance feature extraction module is used to extract the physical appearance parameters describing the chemical products in the production video stream, and the deep appearance feature extraction module is used to extract the non-physical appearance parameters describing the chemical products in the production video stream.

[0024] In some embodiments of the present application, the specific process of inputting a production video stream into a pre-trained AI visual inspection model and outputting multiple appearance feature parameters corresponding to the production video stream for reflecting the appearance of chemical products includes: using a basic appearance feature extraction module to extract the physical appearance parameters describing the chemical products in the production video stream to obtain original appearance feature parameters; using a deep appearance feature extraction module to extract the non-physical appearance parameters describing the chemical products in the production video stream to obtain deep appearance feature parameters; and using the original appearance feature parameters and the deep appearance feature parameters as multiple appearance feature parameters corresponding to the production video stream for reflecting the appearance of chemical products.

[0025] The basic appearance feature extraction module is used to extract directly observable physical appearance attributes of chemical products from production video streams. These attributes can be directly observed and measured, such as color and shape. The deep appearance feature extraction module is used to extract non-observable physical appearance attributes of chemical products from production video streams. These attributes cannot be directly observed and measured, such as the product's complex texture, gloss, and transparency.

[0026] In some embodiments of the present application, the AI ​​model's basic appearance feature extraction module analyzes the video stream to extract the product's physical appearance parameters, such as color and shape, to obtain raw appearance feature parameters. The AI ​​model's deep appearance feature extraction module further analyzes the video stream to extract the product's non-physical appearance parameters, such as texture and gloss, to obtain deep appearance feature parameters. The raw appearance feature parameters and deep appearance feature parameters are integrated to form a set of multiple appearance feature parameters that comprehensively reflect the appearance of the chemical product.

[0027] Among them, the data table of physical appearance parameters is as follows Figure 5A As shown, the data table of non-physical appearance parameters is as follows Figure 5B shown.

[0028] In some embodiments of the present application, the specific process of generating a pre-trained AI visual inspection model includes: obtaining a sample production video stream of the production environment within a preset period; obtaining each sample key frame from the sample production video stream for reflecting the key visual information of the reactor, the mid-term reflection, the moment of material feeding, and the appearance of the chemical product; for each sample key frame, marking the physical appearance parameters about color and shape and the non-physical appearance parameters about texture and gloss to obtain a model training sample; using an appearance parameter recognition algorithm to construct a basic appearance feature extraction module; using a neural network algorithm to construct a deep appearance feature extraction module; building an AI visual inspection model based on the basic appearance feature extraction module and the deep appearance feature extraction module; performing machine learning on the AI ​​visual inspection model based on the model training sample to obtain a pre-trained AI visual inspection model.

[0029] Specifically, the AI ​​visual detection model is subjected to machine learning based on the model training samples to obtain a pre-trained AI visual detection model. The specific process includes: inputting the model training samples into the AI ​​visual detection model and outputting the model loss value; when the model loss value reaches the minimum, generating a pre-trained AI visual detection model; or, when the model loss value does not reach the minimum, updating the model parameters of the AI ​​visual detection model and continuing to execute the step of inputting the model training samples into the AI ​​visual detection model until the model loss value reaches the minimum.

[0030] S103: Analyze, based on the multiple appearance feature parameters, production process parameters in the production equipment in the production environment that do not meet the preset product production conditions; the production process parameters that do not meet the preset product production conditions are traced back based on a preset process knowledge-oriented graph, which is used to record the correlation between the appearance process mark and the production process parameter set of the production equipment in the production environment; Among them, appearance characteristic parameters refer to parameters extracted from production video streams and used to reflect the appearance characteristics of chemical products, including physical and non-physical appearance parameters. The production environment is the manufacturing site of chemical products, including all equipment, tools and conditions. Production equipment is the machines and devices used to manufacture chemical products in the production environment. Preset product production conditions are pre-defined production conditions and standards to ensure product quality. The preset process knowledge-oriented diagram is a chart used to record and analyze the correlation between production process parameters and product appearance characteristics. Appearance process marking refers to the label or logo of the product appearance characteristics related to the production process parameters.

[0031] The plurality of appearance feature parameters include an appearance process mark and an appearance feature quantization value. The appearance feature quantization value is a specific numerical value of an appearance feature parameter, such as the brightness of a color.

[0032] In some embodiments of the present application, the specific process of analyzing the production process parameters that do not meet the preset product production conditions in the production equipment in the production environment based on multiple appearance feature parameters includes: according to the appearance process mark, obtaining the corresponding appearance feature quantization threshold range from the pre-established mapping relationship between the appearance process mark and the appearance feature quantization threshold range; comparing the appearance feature quantization value with the appearance feature quantization threshold range, and determining the appearance process mark corresponding to the appearance feature quantization value that is not in the appearance feature quantization threshold range as multiple abnormal appearance process marks; tracing the production process parameters of each abnormal appearance process mark from the preset process knowledge-oriented graph to obtain multiple candidate production process parameters; comparing the current value of each candidate production process parameter and its preset standard range to determine the production process parameter in the production equipment in the production environment that does not meet the preset product generation conditions.

[0033] The range of quantitative values ​​for appearance characteristics refers to the preset acceptable range of quantitative values ​​for appearance characteristics used to judge product quality. Abnormal appearance process marks are appearance process marks that exceed the quantitative value range and may affect product quality. Production process parameter traceability is the process of searching for production process parameters related to abnormal appearance process marks within a preset process knowledge-based graph. Candidate production process parameters are production process parameters that require adjustment to correct problems in the production process. The current value is the actual measured value of the production process parameter during the production process. The preset standard range is the ideal range within which the production process parameter should be maintained to ensure product quality.

[0034] In some embodiments of the present application, the specific process of tracing the production process parameters of each abnormal appearance process mark from the preset process knowledge-oriented graph to obtain multiple candidate production process parameters includes: taking each abnormal appearance process mark as a traceability condition, traversing and searching for alternative root nodes that meet the traceability conditions from the preset process knowledge-oriented graph; obtaining all child nodes on the alternative root node; taking the union of all child nodes to obtain multiple candidate production process parameters.

[0035] In one possible implementation, an AI visual inspection model processes production video streams from 5G explosion-proof cameras. The extracted appearance feature parameters include, for example, color uniformity of 85 points, glossiness of 45 points, and the number of surface defects of 6. The pre-established mapping relationship between appearance process markers and quantitative threshold ranges for appearance features is as follows: color uniformity: 80-100 points pass, 60-79 points warning, below 60 points abnormal; glossiness: 70-100 points pass, 40-69 points warning, below 40 points abnormal; number of surface defects: 0-2 pass, 3-5 warning, above 5 abnormal). In this case, a color uniformity score of 85 is marked as abnormal, and a number of 6 surface defects is marked as abnormal. The pre-set process knowledge-based map shows that color uniformity abnormality is related to "stirring speed" and "mixing time." In this case, "stirring speed" and "mixing time" are candidate production process parameters.

[0036] In an embodiment of the present application, the specific process of generating a preset process knowledge-oriented graph includes: obtaining a historical production video stream of a production environment within a preset period; synchronizing each key frame in the historical production video stream for reflecting the key visual information of the reactor in the middle stage, the moment of material feeding, and the appearance of the chemical product with the first production process parameter set at the corresponding moment according to the size of the timestamp; inputting each key frame into a pre-trained AI visual detection model to separate multiple first appearance feature quantization values; mining the correlation relationship between each first appearance feature quantization value and the first production process parameter set; and establishing a preset process knowledge-oriented graph based on the correlation relationship.

[0037] The mid-stage of the reaction kettle's reaction is the intermediate stage of the state of matter in the reactor during the chemical reaction process. The moment of material feeding is the moment when the raw materials are put into the reactor or other production equipment during the production process. The appearance of chemical products refers to the external characteristics of the final product, such as color, shape, texture, etc. The first production process parameter set is the initial production process parameters directly related to the production process, such as temperature, pressure, stirring speed, etc. The first appearance feature quantitative value is quantitative data related to the product appearance extracted from the video, such as color saturation, shape and size. The correlation relationship is the logical connection between the first appearance feature quantitative value and the production process parameters.

[0038] In an embodiment of the present application, by presetting a process knowledge-oriented map, production process parameters related to product appearance quality can be quickly identified and adjusted, thereby improving production efficiency.

[0039] In some embodiments of the present application, the specific process of mining the correlation relationship between each first appearance feature quantization value and the first production process parameter set includes: analyzing the linear relationship coefficient and the nonlinear monotonic relationship coefficient between each first appearance feature quantization value and each first production process parameter in the set; screening out the first production process parameters whose absolute value of the linear relationship coefficient is greater than the first preset threshold and whose nonlinear monotonic relationship coefficient is less than the second preset threshold, as the production process parameters that are correlated with each first appearance feature quantization value, and the preset multiple value of the second preset threshold is equal to the first preset threshold; binding the production process parameters that are correlated with each first appearance feature quantization value and each first appearance feature quantization value to obtain the correlation relationship between each first appearance feature quantization value and the first production process parameter set.

[0040] For example, the linear relationship coefficient can be a Pearson correlation coefficient, which is used to measure the linear correlation between two continuous variables (such as the relationship between temperature and color value). The nonlinear monotonic relationship coefficient can be a Spearman rank correlation coefficient, which measures the monotonic relationship between two variables (such as the nonlinear but consistent change trend between stirring speed and bubble density). The first preset threshold is preferably 0.5, and the second preset threshold is preferably 0.05.

[0041] The correlation relationship includes production process parameters that are correlated with each first appearance feature quantization value.

[0042] In some embodiments of the present application, the specific process of establishing a preset process knowledge-oriented graph based on the correlation relationship includes: assigning a first appearance process tag to each first appearance feature quantization value; treating the first appearance process tag assigned to each first appearance feature quantization value as a root node, and treating the production process parameters that are correlated with each first appearance feature quantization value as multiple child nodes; establishing a knowledge-oriented graph of the root node and multiple child nodes; traversing all root nodes, and identifying shared child nodes as nested nodes from the knowledge-oriented graph; and nesting and linking the established knowledge-oriented graphs based on the nested nodes to obtain a preset process knowledge-oriented graph.

[0043] Among them, the knowledge-oriented graph of the root node and multiple child nodes is as follows: Figure 6 As shown. Some of the contents in the preset process knowledge-oriented diagram are as follows. Figure 7 shown.

[0044] S104, adjusting the parameter values ​​of the production process parameters that do not meet the preset product production conditions according to the preset step size.

[0045] In some embodiments of the present application, the preset step size for color uniformity improvement can be 5 units per adjustment. The preset step size for gloss improvement can be 5 units per adjustment. For example, if the current quantized value is 75, which is lower than the preset acceptable value of 80, the adjustment step size is calculated as (80 - 75) / 5 = 1. The stirring speed is adjusted to 300 rpm + 1 × 5 = 305 rpm.

[0046] In an embodiment of the present application, on the one hand, the pre-trained AI visual inspection model can extract multiple appearance feature parameters, including physical appearance parameters (such as color, shape) and non-physical appearance parameters (such as texture, gloss) from the production video stream collected by the 5G explosion-proof camera. These parameters can comprehensively reflect the appearance information of chemical products and provide more accurate and detailed data support for quality control. By analyzing these appearance feature parameters, any slight changes in the appearance of the product can be discovered in a timely manner, which can meet the high requirements for quality control. On the other hand, by obtaining multiple appearance feature parameters reflecting the appearance of chemical products in real time, these parameters are associated with the set of production process parameters recorded in the preset process knowledge-oriented graph to trace the non-compliant production process parameters. The system can automatically adjust the values ​​of these parameters to adapt to the current production conditions. The dynamic online adjustment mechanism can promptly correct deviations in the production process, reduce product quality problems caused by improper parameter settings, and thus significantly reduce the probability of product errors.

[0047] See Figure 8 , provides a flow chart of a model training method for an AI visual detection model according to an embodiment of the present application. Figure 8 As shown, the method of the embodiment of the present application may include the following steps: S201, obtaining a sample production video stream of a production environment within a preset period; In some embodiments of the present application, an explosion-proof 4K industrial camera (such as Basler ace acA2000-50gc) is used to capture video of the entire process from feeding to discharging in the reactor at 30fps for 7 days (covering different batches and working conditions).

[0048] S202, obtaining each sample key frame from the sample production video stream, which is used to reflect key visual information of the reactor in the middle stage, the moment of material feeding, and the appearance of the chemical product; In some embodiments of the present application, the feeding moment can be the 2nd second after the material enters the reactor (the trigger pressure sensor signal is intercepted synchronously), the mid-reaction stage can be 5 consecutive frames when the temperature reaches the set value (150℃±5℃), and the discharging stage can be a video of the surface flow state of the product when it flows out (the last 3 seconds are intercepted).

[0049] S203, for each sample key frame, annotating the physical appearance parameters of color and shape and the non-physical appearance parameters of texture and gloss to obtain a model training sample; In some examples of this application, an OpenCV color detection script was used to determine the RGB mean value (Lab color space) of the central region of the reaction product as the physical appearance parameter of color. LabelMe polygon annotation was used to annotate the number and diameter (unit: mm) of bubbles as the physical appearance parameter of shape. Experts graded the texture using the ASTM D7869 standard (0-5: smooth → severely cracked). Reflectance was measured using a GL-200 gloss meter with simultaneous calibration, providing a non-physical appearance parameter of gloss.

[0050] S204, using an appearance parameter recognition algorithm to construct a basic appearance feature extraction module; Among them, OpenCV+Pytorch can realize the construction of basic appearance feature extraction module.

[0051] S205, using a neural network algorithm to build a deep appearance feature extraction module; Among them, the deep appearance feature extraction module can use EfficientNet-B3, which is suitable for edge computing, as the backbone network.

[0052] S206, building an AI visual inspection model based on the basic appearance feature extraction module and the deep appearance feature extraction module; In some embodiments of the present application, the basic appearance feature extraction module is integrated into the deep appearance feature extraction module to obtain an AI visual detection model.

[0053] S207: Perform machine learning on the AI ​​visual detection model based on the model training samples to obtain a pre-trained AI visual detection model.

[0054] In an embodiment of the present application, on the one hand, the pre-trained AI visual inspection model can extract multiple appearance feature parameters, including physical appearance parameters (such as color, shape) and non-physical appearance parameters (such as texture, gloss) from the production video stream collected by the 5G explosion-proof camera. These parameters can comprehensively reflect the appearance information of chemical products and provide more accurate and detailed data support for quality control. By analyzing these appearance feature parameters, any slight changes in the appearance of the product can be discovered in a timely manner, which can meet the high requirements for quality control. On the other hand, by obtaining multiple appearance feature parameters reflecting the appearance of chemical products in real time, these parameters are associated with the set of production process parameters recorded in the preset process knowledge-oriented graph to trace the non-compliant production process parameters. The system can automatically adjust the values ​​of these parameters to adapt to the current production conditions. The dynamic online adjustment mechanism can promptly correct deviations in the production process, reduce product quality problems caused by improper parameter settings, and thus significantly reduce the probability of product errors.

[0055] The following are device embodiments of the present application, which can be used to implement the method embodiments of the present application. For details not disclosed in the device embodiments of the present application, please refer to the method embodiments of the present application.

[0056] See Figure 9 , which shows a schematic diagram of the structure of a dynamic adjustment device for production process parameters based on AI visual inspection, provided by an exemplary embodiment of the present application. This dynamic adjustment device for production process parameters based on AI visual inspection can be implemented as all or part of an electronic device through software, hardware, or a combination of both. The device 1 includes a production video stream acquisition module 10, an appearance feature parameter output module 20, a production process parameter analysis module 30, and a parameter value adjustment module 40.

[0057] The production video stream acquisition module 10 is used to obtain the production video stream collected by the 5G explosion-proof camera pre-deployed in the production environment within a preset time period. The production video stream is used to represent the internal information of the reactor, the material feeding information, and the appearance information of the chemical product; The appearance feature parameter output module 20 is used to input the production video stream into the pre-trained AI visual inspection model and output multiple appearance feature parameters corresponding to the production video stream that are used to reflect the appearance of the chemical product; The production process parameter analysis module 30 is used to analyze, based on multiple appearance feature parameters, production process parameters in production equipment in the production environment that do not meet the preset product production conditions; the production process parameters that do not meet the preset product production conditions are traced back based on a preset process knowledge-oriented graph, which is used to record the correlation between the appearance process mark and the production process parameter set of the production equipment in the production environment; The parameter value adjustment module 40 is used to adjust the parameter values ​​of the production process parameters that do not meet the preset product production conditions according to the preset step size.

[0058] It should be noted that the dynamic adjustment device for production process parameters based on AI visual inspection provided in the above embodiment only uses the division of the above-mentioned functional modules as an example when executing the dynamic adjustment method for production process parameters based on AI visual inspection. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the equipment can be divided into different functional modules to complete all or part of the functions described above. In addition, the dynamic adjustment device for production process parameters based on AI visual inspection provided in the above embodiment and the dynamic adjustment method for production process parameters based on AI visual inspection belong to the same concept. The implementation process thereof is detailed in the method embodiment and will not be repeated here.

[0059] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0060] In an embodiment of the present application, on the one hand, the pre-trained AI visual inspection model can extract multiple appearance feature parameters, including physical appearance parameters (such as color, shape) and non-physical appearance parameters (such as texture, gloss) from the production video stream collected by the 5G explosion-proof camera. These parameters can comprehensively reflect the appearance information of chemical products and provide more accurate and detailed data support for quality control. By analyzing these appearance feature parameters, any slight changes in the appearance of the product can be discovered in a timely manner, which can meet the high requirements for quality control. On the other hand, by obtaining multiple appearance feature parameters reflecting the appearance of chemical products in real time, these parameters are associated with the set of production process parameters recorded in the preset process knowledge-oriented graph to trace the non-compliant production process parameters. The system can automatically adjust the values ​​of these parameters to adapt to the current production conditions. The dynamic online adjustment mechanism can promptly correct deviations in the production process, reduce product quality problems caused by improper parameter settings, and thus significantly reduce the probability of product errors.

[0061] The present application also provides a computer-readable medium having program instructions stored thereon, which, when executed by a processor, implement the method for dynamically adjusting production process parameters based on AI visual inspection provided by the above-mentioned various method embodiments.

[0062] The present application also provides a computer program product containing instructions, which, when run on a computer, enables the computer to execute the method for dynamically adjusting production process parameters based on AI visual inspection of the above-mentioned various method embodiments.

[0063] See Figure 10 , is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 10As shown, the electronic device 1000 may include: at least one processor 1001 , at least one network interface 1004 , a user interface 1003 , a memory 1005 , and at least one communication bus 1002 .

[0064] The communication bus 1002 is used to implement the connection and communication between these components.

[0065] The user interface 1003 may include a display screen (Display) and a camera (Camera). Optionally, the user interface 1003 may also include a standard wired interface and a wireless interface.

[0066] The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).

[0067] The processor 1001 may include one or more processing cores. The processor 1001 utilizes various interfaces and circuits to connect various components within the electronic device 1000. It executes instructions, programs, code sets, or instruction sets stored in the memory 1005, and accesses data stored in the memory 1005 to perform various functions and process data within the electronic device 1000. Optionally, the processor 1001 may be implemented in hardware using at least one of a digital signal processing (DSP), a field-programmable gate array (FPGA), and a programmable logic array (PLA). The processor 1001 may integrate one or a combination of a central processing unit (CPU), a graphics processing unit (GPU), and a modem. The CPU primarily processes the operating system, user interface, and application programs; the GPU is responsible for rendering and drawing content displayed on the display; and the modem handles wireless communications. It is understood that the modem may also be implemented independently of the processor 1001 and implemented on a separate chip.

[0068] Among them, the memory 1005 may include a random access memory (RAM) or a read-only memory (Read-Only Memory). Optionally, the memory 1005 includes a non-transitory computer-readable storage medium. The memory 1005 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 1005 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store data involved in the above-mentioned various method embodiments, etc. The memory 1005 may also be optionally at least one storage system located away from the aforementioned processor 1001. As Figure 10 As shown, the memory 1005 as a computer storage medium may include an operating system, a network communication module, a user interface module, and an application for dynamically adjusting production process parameters based on AI visual inspection.

[0069] exist Figure 10 In the electronic device 1000 shown, the user interface 1003 is mainly used to provide an input interface for the user and obtain user input data; and the processor 1001 can be used to call the production process parameter dynamic adjustment application based on AI visual inspection stored in the memory 1005 and specifically perform the following operations: The production video stream collected by the 5G explosion-proof camera pre-deployed in the production environment is obtained according to the preset cycle. The production video stream is used to represent the internal information of the reactor, material feeding information, and the appearance information of the chemical product; Input the production video stream into the pre-trained AI visual inspection model, and output multiple appearance feature parameters corresponding to the production video stream to reflect the appearance of chemical products; Analyze the production process parameters of the production equipment in the production environment that do not meet the preset product production conditions based on multiple appearance feature parameters; the production process parameters that do not meet the preset product production conditions are traced back based on the preset process knowledge-oriented graph, which is used to record the correlation between the appearance process marks and the production process parameter sets of the production equipment in the production environment; Adjust the parameter values ​​of production process parameters that do not meet the preset product production conditions according to the preset step size.

[0070] In one embodiment, when the processor 1001 analyzes the production process parameters of the production equipment in the production environment that do not meet the preset product production conditions based on the multiple appearance feature parameters, the processor 1001 specifically performs the following operations: According to the appearance process mark, obtaining the corresponding appearance feature quantization threshold range from a pre-established mapping relationship between the appearance process mark and the appearance feature quantization threshold range; comparing the appearance feature quantization value with the appearance feature quantization threshold range, and determining appearance process marks corresponding to the appearance feature quantization value that is not within the appearance feature quantization threshold range as a plurality of abnormal appearance process marks; From the preset process knowledge-oriented graph, the production process parameters of each abnormal appearance process mark are traced to obtain multiple candidate production process parameters; Compare the current value of each candidate production process parameter with its preset standard range to determine the production process parameters in the production equipment of the production environment that do not meet the preset product generation conditions.

[0071] In one embodiment, when executing the generation of the preset process knowledge-oriented graph, the processor 1001 specifically performs the following operations: Obtain historical production video streams of the production environment within a preset period; Synchronize each key frame in the historical production video stream that reflects key visual information such as the middle stage of the reactor, the moment of material feeding, and the appearance of the chemical product with the first production process parameter set at the corresponding moment according to the size of the timestamp; Input each keyframe into a pre-trained AI visual detection model to separate multiple first appearance feature quantization values; Mining a correlation relationship between each first appearance feature quantization value and a first production process parameter set; Based on the correlation relationship, a preset process knowledge-oriented diagram is established.

[0072] In one embodiment, when mining the correlation between each first appearance feature quantized value and the first production process parameter set, the processor 1001 specifically performs the following operations: Analyzing the linear relationship coefficient and the nonlinear monotonic relationship coefficient between each first appearance feature quantization value and each first production process parameter in the set; Screening out a first production process parameter having an absolute value of a linear relationship coefficient greater than a first preset threshold and a nonlinear monotonic relationship coefficient less than a second preset threshold as the production process parameter having a correlation with each first appearance feature quantized value, wherein a preset multiple value of the second preset threshold is equal to the first preset threshold; A production process parameter correlated with each first appearance feature quantized value and each first appearance feature quantized value are bound to obtain a correlation relationship between each first appearance feature quantized value and the first production process parameter set.

[0073] In one embodiment, when the processor 1001 establishes a preset process knowledge-oriented graph based on the correlation relationship, the processor 1001 specifically performs the following operations: assigning a first appearance process mark to each first appearance feature quantized value; The first appearance process mark assigned to each first appearance feature quantization value is set as a root node, and the production process parameters associated with each first appearance feature quantization value are set as multiple child nodes; Establish a knowledge-oriented graph with a root node and multiple child nodes; Traverse all root nodes and identify shared child nodes as nested nodes from the knowledge-oriented graph; Based on the nested nodes, each knowledge-oriented graph established is nested and linked to obtain a preset process knowledge-oriented graph.

[0074] In one embodiment, when the processor 1001 performs production process parameter tracing for each abnormal appearance process mark from a preset process knowledge-oriented graph and obtains multiple candidate production process parameters, the processor 1001 specifically performs the following operations: Taking each abnormal appearance process mark as the traceability condition, traverse and search for the candidate root node that meets the traceability condition from the preset process knowledge-oriented graph; Get all child nodes on the alternative root node; Take the union of all child nodes to obtain multiple candidate production process parameters.

[0075] In one embodiment, when the processor 1001 inputs the production video stream into the pre-trained AI visual inspection model and outputs multiple appearance feature parameters corresponding to the production video stream and used to reflect the appearance of the chemical product, the processor 1001 specifically performs the following operations: The basic appearance feature extraction module is used to extract the physical appearance parameters describing chemical products from the production video stream to obtain the original appearance feature parameters; A deep appearance feature extraction module is used to extract non-physical appearance parameters describing chemical products from production video streams to obtain deep appearance feature parameters. The original appearance feature parameters and the deep appearance feature parameters are used as multiple appearance feature parameters corresponding to the production video stream and used to reflect the appearance of the chemical product.

[0076] In one embodiment, the processor 1001 performs the following operations when generating a pre-trained AI visual detection model: Obtain sample production video streams of the production environment within a preset period; From the sample production video stream, obtain each sample key frame used to reflect the key visual information of the reactor in the middle of the reaction, the moment of material feeding, and the appearance of the chemical product; For each sample keyframe, the physical appearance parameters of color and shape and the non-physical appearance parameters of texture and gloss are annotated to obtain the model training sample; Adopt appearance parameter recognition algorithm to build basic appearance feature extraction module; Adopt neural network algorithm to build deep appearance feature extraction module; Build an AI visual inspection model based on the basic appearance feature extraction module and the deep appearance feature extraction module; Based on the model training samples, machine learning is performed on the AI ​​visual detection model to obtain a pre-trained AI visual detection model.

[0077] In one embodiment, the processor 1001 performs the following operations when performing machine learning on the AI ​​visual detection model based on the model training samples to obtain a pre-trained AI visual detection model: Input the model training samples into the AI ​​visual detection model and output the model loss value; When the model loss value reaches the minimum, a pre-trained AI visual detection model is generated; or, when the model loss value does not reach the minimum, the model parameters of the AI ​​visual detection model are updated, and the step of inputting the model training samples into the AI ​​visual detection model is continued until the model loss value reaches the minimum.

[0078] In an embodiment of the present application, on the one hand, the pre-trained AI visual inspection model can extract multiple appearance feature parameters, including physical appearance parameters (such as color, shape) and non-physical appearance parameters (such as texture, gloss) from the production video stream collected by the 5G explosion-proof camera. These parameters can comprehensively reflect the appearance information of chemical products and provide more accurate and detailed data support for quality control. By analyzing these appearance feature parameters, any slight changes in the appearance of the product can be discovered in a timely manner, which can meet the high requirements for quality control. On the other hand, by obtaining multiple appearance feature parameters reflecting the appearance of chemical products in real time, these parameters are associated with the set of production process parameters recorded in the preset process knowledge-oriented graph to trace the non-compliant production process parameters. The system can automatically adjust the values ​​of these parameters to adapt to the current production conditions. The dynamic online adjustment mechanism can promptly correct deviations in the production process, reduce product quality problems caused by improper parameter settings, and thus significantly reduce the probability of product errors.

[0079] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware through a computer program. The program for dynamically adjusting production process parameters based on AI visual inspection can be stored in a computer-readable storage medium. When executed, the program can include the processes in the above-described method embodiments. The storage medium for the program for dynamically adjusting production process parameters based on AI visual inspection can be a magnetic disk, an optical disk, a read-only memory, or a random access memory.

[0080] The above disclosure is only a preferred embodiment of the present application, and certainly cannot be used to limit the scope of rights of the present application. Therefore, equivalent changes made according to the claims of the present application are still within the scope covered by the present application.

Claims

1. A method for dynamically adjusting production process parameters based on AI visual inspection, characterized in that: Applied to the server, the method includes: The production video stream collected by the 5G explosion-proof camera pre-deployed in the production environment is obtained according to the preset cycle. The production video stream is used to represent the internal information of the reactor, material feeding information, and the appearance information of the chemical product; Inputting the production video stream into a pre-trained AI visual inspection model, and outputting a plurality of appearance feature parameters corresponding to the production video stream and used to reflect the appearance of the chemical product; analyzing, based on the multiple appearance feature parameters, production process parameters in the production equipment in the production environment that do not meet the preset product production conditions; the production process parameters that do not meet the preset product production conditions are traced back based on a preset process knowledge-oriented graph, which is used to record the correlation relationship between the appearance process mark and the set of production process parameters of the production equipment in the production environment; Adjust the parameter values ​​of production process parameters that do not meet the preset product production conditions according to the preset step size.

2. The method according to claim 1, characterized in that Each appearance feature parameter includes an appearance process mark and an appearance feature quantitative value; The analyzing, based on the multiple appearance characteristic parameters, the production process parameters of the production equipment in the production environment that do not meet the preset product production conditions includes: According to the appearance process mark, obtaining a corresponding appearance feature quantization threshold range from a pre-established mapping relationship between the appearance process mark and the appearance feature quantization threshold range; comparing the appearance feature quantization value with the appearance feature quantization threshold range, and determining appearance process marks corresponding to the appearance feature quantization values ​​that are not within the appearance feature quantization threshold range as a plurality of abnormal appearance process marks; From the preset process knowledge-oriented graph, the production process parameters of each abnormal appearance process mark are traced to obtain multiple candidate production process parameters; The current value of each candidate production process parameter and its preset standard range are compared to determine the production process parameters in the production equipment of the production environment that do not meet the preset product generation conditions.

3. The method according to claim 1, characterized in that The following steps are used to generate a preset process knowledge-oriented diagram, including: Obtaining historical production video streams of the production environment within a preset period; Synchronize each key frame in the historical production video stream, which is used to reflect key visual information of the reactor during the middle phase, the moment of material feeding, and the appearance of the chemical product, with the first production process parameter set at the corresponding moment according to the size of the timestamp; Inputting each key frame into a pre-trained AI visual detection model to separate a plurality of first appearance feature quantization values; Mining a correlation relationship between each first appearance feature quantization value and the first production process parameter set; Based on the correlation relationship, a preset process knowledge-oriented graph is established.

4. The method according to claim 3, characterized in that The mining of the correlation relationship between each first appearance feature quantization value and the first production process parameter set includes: Analyzing the linear relationship coefficient and the nonlinear monotonic relationship coefficient between each first appearance feature quantization value and each first production process parameter in the set; Screening out a first production process parameter whose absolute value of the linear relationship coefficient is greater than a first preset threshold and whose nonlinear monotonic relationship coefficient is less than a second preset threshold as the production process parameter correlated with each of the first appearance feature quantized values, wherein a preset multiple value of the second preset threshold is equal to the first preset threshold; Binding a production process parameter correlated with each first appearance feature quantized value and each first appearance feature quantized value to obtain a correlation relationship between each first appearance feature quantized value and the first production process parameter set.

5. The method according to claim 3, characterized in that The correlation relationship includes production process parameters that are correlated with each of the first appearance feature quantized values; The step of establishing a preset process knowledge-oriented graph based on the correlation relationship includes: assigning a first appearance process mark to each of the first appearance feature quantized values; The first appearance process mark assigned to each first appearance feature quantization value is set as a root node, and the production process parameters associated with each first appearance feature quantization value are set as multiple child nodes; Establishing a knowledge-oriented graph of the root node and the plurality of child nodes; Traversing all root nodes, identifying shared child nodes as nested nodes from the knowledge-oriented graph; Based on the nested nodes, each established knowledge-oriented graph is nested and linked to obtain a preset process knowledge-oriented graph.

6. The method according to claim 2, characterized in that The production process parameters of each abnormal appearance process mark are traced from the preset process knowledge-oriented graph to obtain multiple candidate production process parameters, including: Taking each abnormal appearance process mark as a traceability condition, traverse and search for an alternative root node that meets the traceability condition from a preset process knowledge-oriented graph; Obtain all child nodes of the candidate root node; A union of all the sub-nodes is taken to obtain a plurality of candidate production process parameters.

7. The method according to any one of claims 1 to 6, characterized in that The pre-trained AI visual inspection model includes a basic appearance feature extraction module and a deep appearance feature extraction module; the basic appearance feature extraction module is used to extract physical appearance parameters describing chemical products in production video streams, and the deep appearance feature extraction module is used to extract non-physical appearance parameters describing chemical products in production video streams; The production video stream is input into a pre-trained AI visual inspection model, and multiple appearance feature parameters corresponding to the production video stream and used to reflect the appearance of the chemical product are output, including: Using the basic appearance feature extraction module, extracting physical appearance parameters describing the chemical product in the production video stream to obtain original appearance feature parameters; Using the deep appearance feature extraction module, extract non-physical appearance parameters describing chemical products in the production video stream to obtain deep appearance feature parameters; The original appearance feature parameters and the depth appearance feature parameters are used as multiple appearance feature parameters corresponding to the production video stream and used to reflect the appearance of the chemical product.

8. The method according to claim 7, characterized in that Follow these steps to generate a pre-trained AI visual inspection model, including: Obtaining a sample production video stream of the production environment within a preset period; Obtaining each sample key frame from the sample production video stream, which is used to reflect key visual information of the reactor during the reaction period, the moment of material feeding, and the appearance of the chemical product; For each of the sample key frames, the physical appearance parameters of color and shape and the non-physical appearance parameters of texture and gloss are marked to obtain a model training sample; Adopt appearance parameter recognition algorithm to build basic appearance feature extraction module; Adopt neural network algorithm to build deep appearance feature extraction module; Building an AI visual detection model based on the basic appearance feature extraction module and the deep appearance feature extraction module; The AI ​​visual detection model is machine-learned based on the model training samples to obtain a pre-trained AI visual detection model.

9. The method according to claim 8, characterized in that Performing machine learning on the AI ​​visual inspection model based on the model training samples to obtain a pre-trained AI visual inspection model, including: Input the model training sample into the AI ​​visual detection model and output the model loss value; When the model loss value reaches the minimum, a pre-trained AI visual detection model is generated; or, when the model loss value does not reach the minimum, the model parameters of the AI ​​visual detection model are updated, and the step of inputting the model training samples into the AI ​​visual detection model is continued until the model loss value reaches the minimum.

10. A dynamic adjustment device for production process parameters based on AI visual inspection, characterized in that: The device comprises: The production video stream acquisition module is used to obtain the production video stream collected by 5G explosion-proof cameras pre-deployed in the production environment within a preset time period. The production video stream is used to represent the internal information of the reactor, material feeding information, and the appearance information of the chemical product; An appearance feature parameter output module, configured to input the production video stream into a pre-trained AI visual inspection model and output a plurality of appearance feature parameters corresponding to the production video stream and used to reflect the appearance of the chemical product; a production process parameter analysis module for analyzing, based on the multiple appearance feature parameters, production process parameters in the production equipment in the production environment that do not meet the preset product production conditions; the production process parameters that do not meet the preset product production conditions are obtained by tracing back based on a preset process knowledge-oriented graph, which is used to record the correlation between the appearance process mark and the set of production process parameters of the production equipment in the production environment; The parameter value adjustment module is used to adjust the parameter values ​​of production process parameters that do not meet the preset product production conditions according to the preset step size.

Images